How we research

1. Define the question before researching

We start with the decision a reader is trying to make. A payment platform for a small Indian business may need a different assessment from one for an international enterprise. A broker for a first-time investor may suit a different audience from an active trader.

The review should identify the market, intended customer, product or plan, research period and comparison criteria. Pricing, availability, support and terms must be considered in that context. We explain important exclusions rather than implying complete coverage.

2. Check multiple source types

Our aim is to compare relevant primary evidence with independent reporting and customer experiences. A collection of pages repeating the same press release is not a collection of independent confirmations. We do not use a fixed source count as a substitute for evidence quality.

Official company information

Product documentation, pricing pages, terms, cancellation and refund policies, support documentation, service-status pages and company announcements help establish what is offered. Where relevant, public filings and regulator records can provide additional context. Marketing claims are attributed to the company until supported independently.

Review platforms

Depending on the category, research may include G2, Capterra, Trustpilot, TrustRadius, app-store reviews and other relevant review sites. We note the platform, product listing, dates and number of reviews examined. A platform's “verified” label is reported as that platform's designation, not as our own verification.

Google Business Profile reviews

Google Business Profile reviews—formerly associated with Google My Business—and Google Maps listings can help assess location-specific service experiences. We check that a listing belongs to the correct business and branch. One branch's feedback should not be presented as evidence about an entire company or an unrelated online product.

Reddit and community discussions

Relevant Reddit communities, specialist forums and public user groups can reveal detailed use cases, recurring questions and reported problems. We read the surrounding discussion, dates, replies and any disclosed affiliations. Anonymous posts remain attributed accounts; popularity, upvotes and confident language do not prove a claim.

Instagram and other public social channels

Public Instagram comments, creator posts, YouTube discussions, LinkedIn and other relevant channels may add context about delivery, product experience or support. We distinguish paid promotions, affiliate content, company responses and customer accounts where disclosures are available. Reposted claims and coordinated activity should not be counted as separate experiences.

Independent reporting and specialist analysis

Credible reporting, documented investigations and specialist research can help explain wider context. We trace important claims to their original evidence where possible, record publication dates and distinguish reporting from commentary.

3. Speak with customers where possible

For reviews where direct experience would materially improve the research, we aim to reach out to actual customers. Outreach is an intended research step, not a claim that interviews have already happened. When interviews are conducted, the article should report how many people responded, how they were recruited and the limits of that sample. If no customers were interviewed, we say so.

Useful questions include what they bought, when and where they used it, which plan or product they chose, what worked, what went wrong, how support responded and whether the issue was resolved. Recent experiences and older experiences are kept distinct.

Where appropriate and with consent, we may request redacted supporting evidence such as an invoice, order confirmation or support correspondence. Readers should not send passwords, account access, payment-card details or unnecessary personal information. Supporting documents can substantiate parts of an account; they do not prove every statement or make one person's experience representative.

We seek permission before publishing private correspondence or identifying an interviewee. Where a source requests anonymity, we explain the relevant context without exposing personal details. Brand-referred customers and any relevant incentives or relationships should be disclosed.

4. Cross-check material claims

We compare important claims against evidence suited to the question. A current fee should be checked against current pricing and terms. A reported service failure may require dated correspondence, a status notice, other independent accounts or a company response. Different types of evidence answer different questions.

For example, several complaints about delayed refunds can justify investigating a pattern. They do not, by themselves, establish how often delays occur across all customers. We check the applicable policy, transaction dates, outcomes and available responses before drawing a conclusion.

When sources disagree, we examine whether they refer to different countries, plans, dates, branches or product versions. We keep unresolved disagreements visible. A serious allegation is attributed and distinguished from an established finding; relevant brands should have a reasonable opportunity to respond. A lack of response is not an admission.

5. Put public feedback in context

Public reviews are a selected sample. People with very positive or negative experiences may be more likely to post, and platforms have different moderation, verification and incentive practices. We do not treat social comments or interviews as a statistically representative survey.

• State the sample: identify the platform, date range, selection approach and number of items examined when summarising feedback.
• Look across sentiment: consider positive, mixed and negative experiences rather than selecting only comments that support a preferred conclusion.
• Check repetition: avoid counting duplicate posts, syndicated material or the same incident multiple times.
• Consider recency: distinguish current issues from older reports that may concern a previous product or policy.
• Use cautious language: suspicious wording or a burst of reviews can limit confidence, but is not proof that a reviewer is fake.
• Keep ratings separate: preserve each platform's scale, review count and access date. We do not average incompatible ratings into an unexplained score.

If we calculate a percentage from a collected sample, we state the numerator, denominator and sampling limitations. “Common in the comments we examined” does not mean “common among all customers.”

6. Separate hands-on testing from desk research

A review is labelled hands-on only when the author has actually used the product and documented the relevant checks. The article should explain the plan or version, device or setup, test period, tasks performed and limitations. A demonstration, a trial account and sustained paid use are different levels of access.

When research relies on documentation, public feedback and interviews, it is identified as desk research. We do not write as though we personally tested a product because someone else did. Screenshots, measurements and claims of first-hand experience must reflect work actually performed.

7. Explain the verdict and the trade-offs

Our conclusions should connect evidence to a reader's use case: who the product may suit, who should consider an alternative, what it costs and which limitations matter. A strong verdict includes meaningful benefits as well as drawbacks.

Comparisons should use equivalent plans, markets and time periods wherever possible. Taxes, introductory offers, contract length, usage limits and add-on charges should be identified when material. “Best” recommendations need stated criteria and a defined audience; they are not universal claims.

We do not assign a numerical score unless the scoring method can be explained. Financial-product coverage evaluates the service and available evidence; it does not turn customer sentiment into a prediction of investment returns.

8. Show our work and keep it current

Each published review should include a research note with:

• The author, original publication date and date of any substantive update.
• The product, plan, geography and period covered.
• The sources consulted, linked where possible, with access dates for time-sensitive evidence.
• The feedback sample and selection method, plus customer interview and testing details if applicable.
• Material gaps, conflicting evidence and claims that could not be verified.
• Relevant affiliate relationships, sponsorship, free access or other commercial interests.

Private interview evidence is not automatically made public. Published attribution and summaries should let readers understand the basis of a claim while respecting agreed privacy boundaries.

We aim to recheck volatile facts before publication and when substantive updates are made. An update date should reflect a meaningful change, not an automatic freshness label. Material corrections should describe what changed and why.

Read our editorial policy, commercial disclosures and corrections process for the standards that accompany this methodology.